Separation of composite defect patterns on wafer bin map using support vector clustering

Chih-Hsuan Wang*

*Corresponding author for this work

Research output: Contribution to journalArticle

24 Scopus citations

Abstract

By continuously employing new technologies, huge capital investment, and well-trained engineers, integrated circuit (IC) companies compete intensively with each other to enhance product yield, reduce manufacturing cost and increase profitability. However, wafer fabrication is a complex, costly, and lengthy process that involves hundreds of chemical steps and needs to monitor lots of process parameters at the same time. As we know, IC chips fabricated on semiconductor wafers are highly vulnerable to clustered defects since defects may cause IC to completely malfunction. In particular, various types of defect patterns (i.e. scratch, ring, or zone pattern) shown on the wafer bin map (WBM) usually contain crucial information for quality engineers to track their root causes of failure. Today, many companies still rely on the visual inspection of experienced experts to check and to hand-mark the defective regions on WBM. This manual approach is not only time-consuming, but also inefficient and inconsistent owing to human fatigue. In this study, a hybrid approach that integrates spatial statistics, kernel based eigen-decomposition and support vector clustering is proposed to estimate the number of defect clusters in advance, and to separate both convex and non-convex defect clusters at the same time. Experimental results confirm that four kinds of composite defect patterns are successfully extracted and separated.

Original languageEnglish
Pages (from-to)2554-2561
Number of pages8
JournalExpert Systems with Applications
Volume36
Issue number2 PART 1
DOIs
StatePublished - 1 Jan 2009

Keywords

  • Data mining
  • Defect pattern
  • Kernel decomposition
  • Support vector clustering
  • Wafer binary map

Fingerprint Dive into the research topics of 'Separation of composite defect patterns on wafer bin map using support vector clustering'. Together they form a unique fingerprint.

  • Cite this